static bool pmi_startup(void) { #if WANT_CRAY_PMI2_EXT int spawned, size, rank, appnum; if (PMI2_Initialized()) { /* already initialized */ return true; } /* if we can't startup PMI, we can't be used */ if (PMI_SUCCESS != PMI2_Init(&spawned, &size, &rank, &appnum)) { return false; } /* flag that we started PMI */ started_by_me = true; /* ignore the info - we'll pick it up elsewhere */ return true; #else PMI_BOOL initialized; if (PMI_SUCCESS != PMI_Initialized(&initialized)) { return false; } if (PMI_TRUE != initialized) { if (PMI_SUCCESS != PMI_Init(&initialized)) { return false; } /* flag that we started PMI */ started_by_me = true; } return true; #endif }
int main(int argc, char **argv, char **envp) { int pmi_rank = -1; int pmi_process_group_size = -1; int rc = EXIT_SUCCESS; char *err = NULL; PMI_BOOL pmi_initialized = PMI_FALSE; int i; double pi; int spawned; if (1 < argc) { rc = strtol(argv[1], NULL, 10); } else { rc = 3; } /* sanity */ if (PMI_SUCCESS != PMI_Initialized(&pmi_initialized) || PMI_TRUE == pmi_initialized) { fprintf(stderr, "=== ERROR: PMI sanity failure\n"); return EXIT_FAILURE; } if (PMI_SUCCESS != PMI_Init(&spawned)) { err = "PMI_Init failure!"; goto done; } if (PMI_SUCCESS != PMI_Get_size(&pmi_process_group_size)) { err = "PMI_Get_size failure!"; goto done; } if (PMI_SUCCESS != PMI_Get_rank(&pmi_rank)) { err = "PMI_Get_rank failure!"; goto done; } i = 0; while (1) { i++; pi = i / 3.14159256; if (i > 10000) i = 0; if ((pmi_rank == 3 || (pmi_process_group_size <= 3 && pmi_rank == 0)) && i == 9995) { asprintf(&err, "RANK%d CALLED ABORT", pmi_rank); fprintf(stderr, "%s\n", err); fflush(stderr); PMI_Abort(rc, err); } } done: if (NULL != err) { fprintf(stderr, "=== ERROR [rank:%d] %s\n", pmi_rank, err); rc = EXIT_FAILURE; } return rc; }
static int test_item2(void) { int rc = 0; PMI_BOOL val; if (PMI_SUCCESS != (rc = PMI_Initialized(&val))) { log_fatal("PMI_Initialized failed: %d\n", rc); return rc; } log_assert(PMI_TRUE == val, ""); return rc; }
int orte_grpcomm_pmi_close(void) { #if WANT_CRAY_PMI2_EXT if (started_by_me && PMI2_Initialized()) { PMI2_Finalize(); } #else PMI_BOOL initialized; /* if we weren't selected, cleanup */ if (started_by_me && PMI_SUCCESS == PMI_Initialized(&initialized) && PMI_TRUE == initialized) { PMI_Finalize(); } #endif return ORTE_SUCCESS; }
int main(int argc, char* argv[]) { int rank; int size; int i = 1; char *kvs_name; int kvs_name_l; MPI_Init(&argc, &argv); MPI_Comm_rank(MPI_COMM_WORLD, &rank); MPI_Comm_size(MPI_COMM_WORLD, &size); PMI_BOOL initialized; int ret = PMI_Initialized( &initialized ); printf("%d: ret=%d, init=%d\n",rank,ret,initialized); if( initialized ){ get_kvs_name(&kvs_name, &kvs_name_l); printf("%d: kvs_name = %s\n",kvs_name); } MPI_Finalize(); return 0; }
static int s1_init(void) { PMI_BOOL initialized; int spawned; int rc, ret = OPAL_ERROR; int i, rank, lrank, nrank; char *pmix_id, tmp[64]; opal_value_t kv; char *str; uint32_t ui32; opal_process_name_t ldr; char **localranks=NULL; if (PMI_SUCCESS != (rc = PMI_Initialized(&initialized))) { OPAL_PMI_ERROR(rc, "PMI_Initialized"); return OPAL_ERROR; } if (PMI_TRUE != initialized && PMI_SUCCESS != (rc = PMI_Init(&spawned))) { OPAL_PMI_ERROR(rc, "PMI_Init"); return OPAL_ERROR; } // setup hash table opal_pmix_base_hash_init(); // Initialize space demands rc = PMI_KVS_Get_value_length_max(&pmix_vallen_max); if (PMI_SUCCESS != rc) { OPAL_PMI_ERROR(rc, "PMI_KVS_Get_value_length_max"); goto err_exit; } pmix_vallen_threshold = pmix_vallen_max * 3; pmix_vallen_threshold >>= 2; rc = PMI_KVS_Get_name_length_max(&pmix_kvslen_max); if (PMI_SUCCESS != rc) { OPAL_PMI_ERROR(rc, "PMI_KVS_Get_name_length_max"); goto err_exit; } rc = PMI_KVS_Get_key_length_max(&pmix_keylen_max); if (PMI_SUCCESS != rc) { OPAL_PMI_ERROR(rc, "PMI_KVS_Get_key_length_max"); goto err_exit; } // Initialize job environment information pmix_id = (char*)malloc(pmix_vallen_max); if (pmix_id == NULL) { ret = OPAL_ERR_OUT_OF_RESOURCE; goto err_exit; } /* Get domain id */ if (PMI_SUCCESS != (rc = PMI_Get_kvs_domain_id(pmix_id, pmix_vallen_max))) { free(pmix_id); goto err_exit; } /* get our rank */ ret = PMI_Get_rank(&rank); if( PMI_SUCCESS != ret ) { OPAL_PMI_ERROR(ret, "PMI_Get_rank"); free(pmix_id); goto err_exit; } /* Slurm PMI provides the job id as an integer followed * by a '.', followed by essentially a stepid. The first integer * defines an overall job number. The second integer is the number of * individual jobs we have run within that allocation. */ s1_pname.jobid = strtoul(pmix_id, &str, 10); s1_pname.jobid = (s1_pname.jobid << 16) & 0xffff0000; if (NULL != str) { ui32 = strtoul(str, NULL, 10); s1_pname.jobid |= (ui32 & 0x0000ffff); } ldr.jobid = s1_pname.jobid; s1_pname.vpid = rank; /* store our name in the opal_proc_t so that * debug messages will make sense - an upper * layer will eventually overwrite it, but that * won't do any harm */ opal_proc_set_name(&s1_pname); opal_output_verbose(2, opal_pmix_base_framework.framework_output, "%s pmix:s1: assigned tmp name", OPAL_NAME_PRINT(s1_pname)); OBJ_CONSTRUCT(&kv, opal_value_t); kv.key = strdup(OPAL_PMIX_JOBID); kv.type = OPAL_UINT32; kv.data.uint32 = s1_pname.jobid; if (OPAL_SUCCESS != (ret = opal_pmix_base_store(&OPAL_PROC_MY_NAME, &kv))) { OPAL_ERROR_LOG(ret); OBJ_DESTRUCT(&kv); goto err_exit; } OBJ_DESTRUCT(&kv); /* save it */ OBJ_CONSTRUCT(&kv, opal_value_t); kv.key = strdup(OPAL_PMIX_RANK); kv.type = OPAL_UINT32; kv.data.uint32 = rank; if (OPAL_SUCCESS != (ret = opal_pmix_base_store(&OPAL_PROC_MY_NAME, &kv))) { OPAL_ERROR_LOG(ret); OBJ_DESTRUCT(&kv); goto err_exit; } OBJ_DESTRUCT(&kv); pmix_kvs_name = (char*)malloc(pmix_kvslen_max); if (pmix_kvs_name == NULL) { ret = OPAL_ERR_OUT_OF_RESOURCE; goto err_exit; } rc = PMI_KVS_Get_my_name(pmix_kvs_name, pmix_kvslen_max); if (PMI_SUCCESS != rc) { OPAL_PMI_ERROR(rc, "PMI_KVS_Get_my_name"); goto err_exit; } /* get our local proc info to find our local rank */ if (PMI_SUCCESS != (rc = PMI_Get_clique_size(&nlranks))) { OPAL_PMI_ERROR(rc, "PMI_Get_clique_size"); return rc; } /* save the local size */ OBJ_CONSTRUCT(&kv, opal_value_t); kv.key = strdup(OPAL_PMIX_LOCAL_SIZE); kv.type = OPAL_UINT32; kv.data.uint32 = nlranks; if (OPAL_SUCCESS != (ret = opal_pmix_base_store(&OPAL_PROC_MY_NAME, &kv))) { OPAL_ERROR_LOG(ret); OBJ_DESTRUCT(&kv); goto err_exit; } OBJ_DESTRUCT(&kv); lrank = 0; nrank = 0; ldr.vpid = rank; if (0 < nlranks) { /* now get the specific ranks */ lranks = (int*)calloc(nlranks, sizeof(int)); if (NULL == lranks) { rc = OPAL_ERR_OUT_OF_RESOURCE; OPAL_ERROR_LOG(rc); return rc; } if (PMI_SUCCESS != (rc = PMI_Get_clique_ranks(lranks, nlranks))) { OPAL_PMI_ERROR(rc, "PMI_Get_clique_ranks"); free(lranks); return rc; } /* note the local ldr */ ldr.vpid = lranks[0]; /* save this */ memset(tmp, 0, 64); for (i=0; i < nlranks; i++) { (void)snprintf(tmp, 64, "%d", lranks[i]); opal_argv_append_nosize(&localranks, tmp); if (rank == lranks[i]) { lrank = i; nrank = i; } } str = opal_argv_join(localranks, ','); opal_argv_free(localranks); OBJ_CONSTRUCT(&kv, opal_value_t); kv.key = strdup(OPAL_PMIX_LOCAL_PEERS); kv.type = OPAL_STRING; kv.data.string = str; if (OPAL_SUCCESS != (ret = opal_pmix_base_store(&OPAL_PROC_MY_NAME, &kv))) { OPAL_ERROR_LOG(ret); OBJ_DESTRUCT(&kv); goto err_exit; } OBJ_DESTRUCT(&kv); } /* save the local leader */ OBJ_CONSTRUCT(&kv, opal_value_t); kv.key = strdup(OPAL_PMIX_LOCALLDR); kv.type = OPAL_UINT64; kv.data.uint64 = *(uint64_t*)&ldr; if (OPAL_SUCCESS != (ret = opal_pmix_base_store(&OPAL_PROC_MY_NAME, &kv))) { OPAL_ERROR_LOG(ret); OBJ_DESTRUCT(&kv); goto err_exit; } OBJ_DESTRUCT(&kv); /* save our local rank */ OBJ_CONSTRUCT(&kv, opal_value_t); kv.key = strdup(OPAL_PMIX_LOCAL_RANK); kv.type = OPAL_UINT16; kv.data.uint16 = lrank; if (OPAL_SUCCESS != (ret = opal_pmix_base_store(&OPAL_PROC_MY_NAME, &kv))) { OPAL_ERROR_LOG(ret); OBJ_DESTRUCT(&kv); goto err_exit; } OBJ_DESTRUCT(&kv); /* and our node rank */ OBJ_CONSTRUCT(&kv, opal_value_t); kv.key = strdup(OPAL_PMIX_NODE_RANK); kv.type = OPAL_UINT16; kv.data.uint16 = nrank; if (OPAL_SUCCESS != (ret = opal_pmix_base_store(&OPAL_PROC_MY_NAME, &kv))) { OPAL_ERROR_LOG(ret); OBJ_DESTRUCT(&kv); goto err_exit; } OBJ_DESTRUCT(&kv); /* get universe size */ ret = PMI_Get_universe_size(&i); if (PMI_SUCCESS != ret) { OPAL_PMI_ERROR(ret, "PMI_Get_universe_size"); goto err_exit; } /* push this into the dstore for subsequent fetches */ OBJ_CONSTRUCT(&kv, opal_value_t); kv.key = strdup(OPAL_PMIX_UNIV_SIZE); kv.type = OPAL_UINT32; kv.data.uint32 = i; if (OPAL_SUCCESS != (ret = opal_pmix_base_store(&OPAL_PROC_MY_NAME, &kv))) { OPAL_ERROR_LOG(ret); OBJ_DESTRUCT(&kv); goto err_exit; } OBJ_DESTRUCT(&kv); /* get job size */ ret = PMI_Get_size(&i); if (PMI_SUCCESS != ret) { OPAL_PMI_ERROR(ret, "PMI_Get_size"); goto err_exit; } OBJ_CONSTRUCT(&kv, opal_value_t); kv.key = strdup(OPAL_PMIX_JOB_SIZE); kv.type = OPAL_UINT32; kv.data.uint32 = i; if (OPAL_SUCCESS != (ret = opal_pmix_base_store(&OPAL_PROC_MY_NAME, &kv))) { OPAL_ERROR_LOG(ret); OBJ_DESTRUCT(&kv); goto err_exit; } OBJ_DESTRUCT(&kv); /* get appnum */ ret = PMI_Get_appnum(&i); if (PMI_SUCCESS != ret) { OPAL_PMI_ERROR(ret, "PMI_Get_appnum"); goto err_exit; } OBJ_CONSTRUCT(&kv, opal_value_t); kv.key = strdup(OPAL_PMIX_APPNUM); kv.type = OPAL_UINT32; kv.data.uint32 = i; if (OPAL_SUCCESS != (ret = opal_pmix_base_store(&OPAL_PROC_MY_NAME, &kv))) { OPAL_ERROR_LOG(ret); OBJ_DESTRUCT(&kv); goto err_exit; } OBJ_DESTRUCT(&kv); /* increment the init count */ ++pmix_init_count; return OPAL_SUCCESS; err_exit: PMI_Finalize(); return ret; }
int main(int argc, char *argv[]) { int initialized, rank, size; int i, max_name_len, max_key_len, max_val_len; char *name, *key, *val; if (PMI_SUCCESS != PMI_Initialized(&initialized)) { return 1; } if (0 == initialized) { if (PMI_SUCCESS != PMI_Init(&initialized)) { return 1; } } if (PMI_SUCCESS != PMI_Get_rank(&rank)) { return 1; } if (PMI_SUCCESS != PMI_Get_size(&size)) { return 1; } printf("Hello, World. I am %d of %d\n", rank, size); if (PMI_SUCCESS != PMI_KVS_Get_name_length_max(&max_name_len)) { return 1; } name = (char*) malloc(max_name_len); if (NULL == name) return 1; if (PMI_SUCCESS != PMI_KVS_Get_key_length_max(&max_key_len)) { return 1; } key = (char*) malloc(max_key_len); if (NULL == key) return 1; if (PMI_SUCCESS != PMI_KVS_Get_value_length_max(&max_val_len)) { return 1; } val = (char*) malloc(max_val_len); if (NULL == val) return 1; if (PMI_SUCCESS != PMI_KVS_Get_my_name(name, max_name_len)) { return 1; } /* put my information */ snprintf(key, max_key_len, "pmi_hello-%lu-test", (long unsigned) rank); snprintf(val, max_val_len, "%lu", (long unsigned) rank); if (PMI_SUCCESS != PMI_KVS_Put(name, key, val)) { return 1; } if (PMI_SUCCESS != PMI_KVS_Commit(name)) { return 1; } if (PMI_SUCCESS != PMI_Barrier()) { return 1; } /* verify everyone's information */ for (i = 0 ; i < size ; ++i) { snprintf(key, max_key_len, "pmi_hello-%lu-test", (long unsigned) i); if (PMI_SUCCESS != PMI_KVS_Get(name, key, val, max_val_len)) { return 1; } if (i != strtol(val, NULL, 0)) { fprintf(stderr, "%d: Error: Expected %d, got %d\n", rank, i, (int) strtol(val, NULL, 0)); return 1; } } PMI_Finalize(); return 0; }
main (int argc, char **argv) { int i, j, rc; int nprocs, procid; int clique_size, *clique_ranks = NULL; char *jobid_ptr, *nprocs_ptr, *procid_ptr; int pmi_rank, pmi_size, kvs_name_len, key_len, val_len; PMI_BOOL initialized; char *key, *val, *kvs_name; struct timeval tv1, tv2; long delta_t; char tv_str[20]; gettimeofday(&tv1, NULL); /* Get process count and our id from environment variables */ jobid_ptr = getenv("SLURM_JOB_ID"); nprocs_ptr = getenv("SLURM_NPROCS"); procid_ptr = getenv("SLURM_PROCID"); if (jobid_ptr == NULL) { printf("WARNING: PMI test not run under SLURM\n"); nprocs = 1; procid = 0; } else if ((nprocs_ptr == NULL) || (procid_ptr == NULL)) { printf("FAILURE: SLURM environment variables not set\n"); exit(1); } else { nprocs = atoi(nprocs_ptr); procid = atoi(procid_ptr); } /* Validate process count and our id */ if ((nprocs < 1) || (nprocs > 9999)) { printf("FAILURE: Invalid nprocs %s\n", nprocs_ptr); exit(1); } if ((procid < 0) || (procid > 9999)) { printf("FAILURE: Invalid procid %s\n", procid_ptr); exit(1); } /* Get process count and size from PMI and validate */ if ((rc = PMI_Init(&i)) != PMI_SUCCESS) { printf("FAILURE: PMI_Init: %d\n", rc); exit(1); } initialized = PMI_FALSE; if ((rc = PMI_Initialized(&initialized)) != PMI_SUCCESS) { printf("FAILURE: PMI_Initialized: %d\n", rc); exit(1); } if (initialized != PMI_TRUE) { printf("FAILURE: PMI_Initialized returned false\n"); exit(1); } if ((rc = PMI_Get_rank(&pmi_rank)) != PMI_SUCCESS) { printf("FAILURE: PMI_Get_rank: %d\n", rc); exit(1); } #if _DEBUG printf("PMI_Get_rank = %d\n", pmi_rank); #endif if ((rc = PMI_Get_size(&pmi_size)) != PMI_SUCCESS) { printf("FAILURE: PMI_Get_size: %d, task %d\n", rc, pmi_rank); exit(1); } #if _DEBUG printf("PMI_Get_size = %d\n", pmi_size); #endif if (pmi_rank != procid) { printf("FAILURE: Rank(%d) != PROCID(%d)\n", pmi_rank, procid); exit(1); } if (pmi_size != nprocs) { printf("FAILURE: Size(%d) != NPROCS(%d), task %d\n", pmi_size, nprocs, pmi_rank); exit(1); } if ((rc = PMI_Get_clique_size(&clique_size)) != PMI_SUCCESS) { printf("FAILURE: PMI_Get_clique_size: %d, task %d\n", rc, pmi_rank); exit(1); } clique_ranks = malloc(sizeof(int) * clique_size); if ((rc = PMI_Get_clique_ranks(clique_ranks, clique_size)) != PMI_SUCCESS) { printf("FAILURE: PMI_Get_clique_ranks: %d, task %d\n", rc, pmi_rank); exit(1); } #if _DEBUG for (i=0; i<clique_size; i++) printf("PMI_Get_clique_ranks[%d]=%d\n", i, clique_ranks[i]); #endif free(clique_ranks); if ((rc = PMI_KVS_Get_name_length_max(&kvs_name_len)) != PMI_SUCCESS) { printf("FAILURE: PMI_KVS_Get_name_length_max: %d, task %d\n", rc, pmi_rank); exit(1); } #if _DEBUG printf("PMI_KVS_Get_name_length_max = %d\n", kvs_name_len); #endif kvs_name = malloc(kvs_name_len); if ((rc = PMI_KVS_Get_my_name(kvs_name, kvs_name_len)) != PMI_SUCCESS) { printf("FAILURE: PMI_KVS_Get_my_name: %d, task %d\n", rc, pmi_rank); exit(1); } #if _DEBUG printf("PMI_KVS_Get_my_name = %s\n", kvs_name); #endif if ((rc = PMI_KVS_Get_key_length_max(&key_len)) != PMI_SUCCESS) { printf("FAILURE: PMI_KVS_Get_key_length_max: %d, task %d\n", rc, pmi_rank); exit(1); } key = malloc(key_len); if ((rc = PMI_KVS_Get_value_length_max(&val_len)) != PMI_SUCCESS) { printf("FAILURE: PMI_KVS_Get_value_length_max: %d, task %d\n", rc, pmi_rank); exit(1); } #if _DEBUG printf("PMI_KVS_Get_value_length_max = %d\n", val_len); #endif val = malloc(val_len); /* Build and set some key=val pairs */ snprintf(key, key_len, "ATTR_1_%d", procid); snprintf(val, val_len, "A%d", procid+OFFSET_1); if ((rc = PMI_KVS_Put(kvs_name, key, val)) != PMI_SUCCESS) { printf("FAILURE: PMI_KVS_Put(%s,%s,%s): %d, task %d\n", kvs_name, key, val, rc, pmi_rank); exit(1); } #if _DEBUG printf("PMI_KVS_Put(%s,%s,%s)\n", kvs_name, key, val); #endif snprintf(key, key_len, "attr_2_%d", procid); snprintf(val, val_len, "B%d", procid+OFFSET_2); if ((rc = PMI_KVS_Put(kvs_name, key, val)) != PMI_SUCCESS) { printf("FAILURE: PMI_KVS_Put(%s,%s,%s): %d, task %d\n", kvs_name, key, val, rc, pmi_rank); exit(1); } #if _DEBUG printf("PMI_KVS_Put(%s,%s,%s)\n", kvs_name, key, val); #endif /* Sync KVS across all tasks */ if ((rc = PMI_KVS_Commit(kvs_name)) != PMI_SUCCESS) { printf("FAILURE: PMI_KVS_Commit: %d, task %d\n", rc, pmi_rank); exit(1); } #if _DEBUG printf("PMI_KVS_Commit completed\n"); #endif if ((rc = PMI_Barrier()) != PMI_SUCCESS) { printf("FAILURE: PMI_Barrier: %d, task %d\n", rc, pmi_rank); exit(1); } #if _DEBUG printf("PMI_Barrier completed\n"); #endif /* Now lets get all keypairs and validate */ for (i=0; i<pmi_size; i++) { snprintf(key, key_len, "ATTR_1_%d", i); if ((rc = PMI_KVS_Get(kvs_name, key, val, val_len)) != PMI_SUCCESS) { printf("FAILURE: PMI_KVS_Get(%s): %d, task %d\n", key, rc, pmi_rank); exit(1); } if ((val[0] != 'A') || ((atoi(&val[1])-OFFSET_1) != i)) { printf("FAILURE: Bad keypair %s=%s, task %d\n", key, val, pmi_rank); exit(1); } #if _DEBUG if ((pmi_size <= 8) && (pmi_rank == 0)) /* limit output */ printf("PMI_KVS_Get(%s,%s) %s\n", kvs_name, key, val); #endif snprintf(key, key_len, "attr_2_%d", i); if ((rc = PMI_KVS_Get(kvs_name, key, val, val_len)) != PMI_SUCCESS) { printf("FAILURE: PMI_KVS_Get(%s): %d, task %d\n", key, rc, pmi_rank); exit(1); } if ((val[0] != 'B') || ((atoi(&val[1])-OFFSET_2) != i)) { printf("FAILURE: Bad keypair %s=%s, task %d\n", key,val, pmi_rank); exit(1); } #if _DEBUG if ((pmi_size <= 8) && (pmi_rank == 1)) /* limit output */ printf("PMI_KVS_Get(%s,%s) %s\n", kvs_name, key, val); #endif } /* use iterator */ if ((rc = PMI_KVS_Iter_first(kvs_name, key, key_len, val, val_len)) != PMI_SUCCESS) { printf("FAILURE: PMI_KVS_iter_first: %d, task %d\n", rc, pmi_rank); exit(1); } for (i=0; ; i++) { if (key[0] == '\0') { if (i != (pmi_size * 2)) { printf("FAILURE: PMI_KVS_iter_next " "cycle count(%d, %d), task %d\n", i, pmi_size, pmi_rank); } break; } #if _DEBUG if ((pmi_size <= 8) && (pmi_rank == 1)) { /* limit output */ printf("PMI_KVS_Iter_next(%s,%d): %s=%s\n", kvs_name, i, key, val); } #endif if ((rc = PMI_KVS_Iter_next(kvs_name, key, key_len, val, val_len)) != PMI_SUCCESS) { printf("FAILURE: PMI_KVS_iter_next: %d, task %d\n", rc, pmi_rank); exit(1); } } /* Build some more key=val pairs */ snprintf(key, key_len, "ATTR_3_%d", procid); snprintf(val, val_len, "C%d", procid+OFFSET_1); if ((rc = PMI_KVS_Put(kvs_name, key, val)) != PMI_SUCCESS) { printf("FAILURE: PMI_KVS_Put(%s,%s,%s): %d, task %d\n", kvs_name, key, val, rc, pmi_rank); exit(1); } #if _DEBUG printf("PMI_KVS_Put(%s,%s,%s)\n", kvs_name, key, val); #endif snprintf(key, key_len, "attr_4_%d", procid); snprintf(val, val_len, "D%d", procid+OFFSET_2); if ((rc = PMI_KVS_Put(kvs_name, key, val)) != PMI_SUCCESS) { printf("FAILURE: PMI_KVS_Put(%s,%s,%s): %d, task %d\n", kvs_name, key, val, rc, pmi_rank); exit(1); } #if _DEBUG printf("PMI_KVS_Put(%s,%s,%s)\n", kvs_name, key, val); #endif /* Sync KVS across all tasks */ if ((rc = PMI_KVS_Commit(kvs_name)) != PMI_SUCCESS) { printf("FAILURE: PMI_KVS_Commit: %d, task %d\n", rc, pmi_rank); exit(1); } #if _DEBUG printf("PMI_KVS_Commit completed\n"); #endif if ((rc = PMI_Barrier()) != PMI_SUCCESS) { printf("FAILURE: PMI_Barrier: %d, task %d\n", rc, pmi_rank); exit(1); } #if _DEBUG printf("PMI_Barrier completed\n"); #endif /* Now lets get some keypairs and validate */ for (i=0; i<pmi_size; i++) { snprintf(key, key_len, "ATTR_1_%d", i); if ((rc = PMI_KVS_Get(kvs_name, key, val, val_len)) != PMI_SUCCESS) { printf("FAILURE: PMI_KVS_Get(%s): %d, task %d\n", key, rc, pmi_rank); exit(1); } if ((val[0] != 'A') || ((atoi(&val[1])-OFFSET_1) != i)) { printf("FAILURE: Bad keypair %s=%s, task %d\n", key, val, pmi_rank); exit(1); } #if _DEBUG if ((pmi_size <= 8) && (pmi_rank == 1)) /* limit output */ printf("PMI_KVS_Get(%s,%s) %s\n", kvs_name, key, val); #endif snprintf(key, key_len, "attr_4_%d", i); if ((rc = PMI_KVS_Get(kvs_name, key, val, val_len)) != PMI_SUCCESS) { printf("FAILURE: PMI_KVS_Get(%s): %d, task %d\n", key, rc, pmi_rank); exit(1); } if ((val[0] != 'D') || ((atoi(&val[1])-OFFSET_2) != i)) { printf("FAILURE: Bad keypair %s=%s, task %d\n", key,val, pmi_rank); exit(1); } #if _DEBUG if ((pmi_size <= 8) && (pmi_rank == 1)) /* limit output */ printf("PMI_KVS_Get(%s,%s) %s\n", kvs_name, key, val); #endif } /* Replicate the very heavy load that MVAPICH2 puts on PMI * This load exceeds that of MPICH2 by a very wide margin */ #if _DEBUG printf("Starting %d iterations each with %d PMI_KVS_Put and \n" " one each PMI_KVS_Commit and KVS_Barrier\n", BARRIER_CNT, PUTS_PER_BARRIER); fflush(stdout); #endif for (i=0; i<BARRIER_CNT; i++) { for (j=0; j<PUTS_PER_BARRIER; j++) { snprintf(key, key_len, "ATTR_%d_%d_%d", i, j, procid); snprintf(val, val_len, "C%d", procid+OFFSET_1); if ((rc = PMI_KVS_Put(kvs_name, key, val)) != PMI_SUCCESS) { printf("FAILURE: PMI_KVS_Put(%s,%s,%s): " "%d, task %d\n", kvs_name, key, val, rc, pmi_rank); exit(1); } } if ((rc= PMI_KVS_Commit(kvs_name)) != PMI_SUCCESS) { printf("FAILURE: PMI_KVS_Commit: %d, task %d\n", rc, pmi_rank); exit(1); } if ((rc = PMI_Barrier()) != PMI_SUCCESS) { printf("FAILURE: PMI_Barrier: %d, task %d\n", rc, pmi_rank); exit(1); } /* Don't bother with PMI_KVS_Get as those are all local * and do not put a real load on srun or the network */ } #if _DEBUG printf("Interative PMI calls successful\n"); #endif /* create new keyspace and test it */ if ((rc = PMI_KVS_Create(kvs_name, kvs_name_len)) != PMI_SUCCESS) { printf("FAILURE: PMI_KVS_Create: %d, task %d\n", rc, pmi_rank); exit(1); } #if _DEBUG printf("PMI_KVS_Create %s\n", kvs_name); #endif if ((rc = PMI_KVS_Put(kvs_name, "KVS_KEY", "KVS_VAL")) != PMI_SUCCESS) { printf("FAILURE: PMI_KVS_Put: %d, task %d\n", rc, pmi_rank); exit(1); } #if _DEBUG printf("PMI_KVS_Put(%s,KVS_KEY,KVS_VAL)\n", kvs_name); #endif if ((rc = PMI_KVS_Get(kvs_name, "KVS_KEY", val, val_len)) != PMI_SUCCESS) { printf("FAILURE: PMI_KVS_Get(%s, KVS_KEY): %d, task %d\n", kvs_name, rc, pmi_rank); exit(1); } #if _DEBUG printf("PMI_KVS_Get(%s,%s) %s\n", kvs_name, "KVS_KEY", val); #endif if ((rc = PMI_KVS_Destroy(kvs_name)) != PMI_SUCCESS) { printf("FAILURE: PMI_KVS_Destroy(%s): %d, task %d\n", kvs_name, rc, pmi_rank); exit(1); } if ((rc = PMI_KVS_Get(kvs_name, "KVS_KEY", val, val_len)) != PMI_ERR_INVALID_KVS) { printf("FAILURE: PMI_KVS_Get(%s, KVS_KEY): %d, task %d\n", kvs_name, rc, pmi_rank); exit(1); } if ((rc = PMI_Finalize()) != PMI_SUCCESS) { printf("FAILURE: PMI_Finalize: %d, task %d\n", rc, pmi_rank); exit(1); } if (_DEBUG || (pmi_rank < 4)) { gettimeofday(&tv2, NULL); delta_t = (tv2.tv_sec - tv1.tv_sec) * 1000000; delta_t += tv2.tv_usec - tv1.tv_usec; snprintf(tv_str, sizeof(tv_str), "usec=%ld", delta_t); printf("PMI test ran successfully, for task %d, %s\n", pmi_rank, tv_str); } if (pmi_rank == 0) { printf("NOTE: All failures reported, "); printf("but only first four successes reported\n"); } exit(0); }
int main(void) { int rc; int rank, size; PMI_BOOL initialized; rc = PMI_Initialized(&initialized); if (rc!=PMI_SUCCESS) PMI_Abort(rc,"PMI_Initialized failed"); if (initialized!=PMI_TRUE) { int spawned; rc = PMI_Init(&spawned); if (rc!=PMI_SUCCESS) PMI_Abort(rc,"PMI_Init failed"); } rc = PMI_Get_rank(&rank); if (rc!=PMI_SUCCESS) PMI_Abort(rc,"PMI_Get_rank failed"); rc = PMI_Get_size(&size); if (rc!=PMI_SUCCESS) PMI_Abort(rc,"PMI_Get_size failed"); printf("rank %d of %d \n", rank, size); int rpn; /* rpn = ranks per node */ rc = PMI_Get_clique_size(&rpn); if (rc!=PMI_SUCCESS) PMI_Abort(rc,"PMI_Get_clique_size failed"); printf("rank %d clique size %d \n", rank, rpn); int * clique_ranks = malloc( rpn * sizeof(int) ); if (clique_ranks==NULL) PMI_Abort(rpn,"malloc failed"); rc = PMI_Get_clique_ranks(clique_ranks, rpn); if (rc!=PMI_SUCCESS) PMI_Abort(rc,"PMI_Get_clique_ranks failed"); for(int i = 0; i<rpn; i++) printf("rank %d clique[%d] = %d \n", rank, i, clique_ranks[i]); int nid; rc = PMI_Get_nid(rank, &nid); if (rc!=PMI_SUCCESS) PMI_Abort(rc,"PMI_Get_nid failed"); printf("rank %d PMI_Get_nid gives nid %d \n", rank, nid); #if OLD rca_mesh_coord_t xyz; rca_get_meshcoord( (uint16_t) nid, &xyz); printf("rank %d rca_get_meshcoord returns (%2u,%2u,%2u)\n", rank, xyz.mesh_x, xyz.mesh_y, xyz.mesh_z); #else // UNTESTED pmi_mesh_coord_t xyz; PMI_Get_meshcoord((uint16_t) nid, &xyz); printf("rank %d PMI_Get_meshcoord returns (%2u,%2u,%2u)\n", rank, xyz.mesh_x, xyz.mesh_y, xyz.mesh_z); #endif fflush(stdout); return 0; }
/* gasneti_bootstrapInit */ int gasneti_bootstrapInit_pmi( int *argc_p, char ***argv_p, gasnet_node_t *nodes_p, gasnet_node_t *mynode_p) { int size, rank; #if USE_PMI2_API int spawned, appnum; if (PMI2_SUCCESS != PMI2_Init(&spawned, &size, &rank, &appnum)) { return GASNET_ERR_NOT_INIT; } #else if (PMI_SUCCESS != PMI_Initialized(&gasneti_pmi_initialized)) { return GASNET_ERR_NOT_INIT; } if (PMI_FALSE == gasneti_pmi_initialized) { int spawned; if (PMI_SUCCESS != PMI_Init(&spawned)) { return GASNET_ERR_NOT_INIT; } } if (PMI_SUCCESS != PMI_Get_rank(&rank)) { gasneti_fatalerror("PMI_Get_rank() failed"); } if (PMI_SUCCESS != PMI_Get_size(&size)) { gasneti_fatalerror("PMI_Get_size() failed"); } #endif *mynode_p = rank; *nodes_p = size; #if USE_PMI2_API max_name_len = 1024; /* XXX: can almost certainly be shorter than this! */ max_key_len = PMI2_MAX_KEYLEN; max_val_len = PMI2_MAX_VALLEN; #else if (PMI_SUCCESS != PMI_KVS_Get_name_length_max(&max_name_len)) { gasneti_fatalerror("PMI_KVS_Get_name_length_max() failed"); } if (PMI_SUCCESS != PMI_KVS_Get_key_length_max(&max_key_len)) { gasneti_fatalerror("PMI_KVS_Get_key_length_max() failed"); } if (PMI_SUCCESS != PMI_KVS_Get_value_length_max(&max_val_len)) { gasneti_fatalerror("PMI_KVS_Get_value_length_max() failed"); } #endif kvs_name = (char*) gasneti_malloc(max_name_len); kvs_key = (char*) gasneti_malloc(max_key_len); kvs_value = (char*) gasneti_malloc(max_val_len); max_val_bytes = 4 * (max_val_len / 5); #if USE_PMI2_API if (PMI2_SUCCESS != PMI2_Job_GetId(kvs_name, max_name_len)) { gasneti_fatalerror("PMI2_Job_GetId() failed"); } #else if (PMI_SUCCESS != PMI_KVS_Get_my_name(kvs_name, max_name_len)) { gasneti_fatalerror("PMI_KVS_Get_my_name() failed"); } #endif return GASNET_OK; }
static int s1_init(void) { PMI_BOOL initialized; int spawned; int rc, ret = OPAL_ERROR; int i; char *pmix_id, *tmp; uint32_t jobfam, stepid; opal_value_t kv; if (PMI_SUCCESS != (rc = PMI_Initialized(&initialized))) { OPAL_PMI_ERROR(rc, "PMI_Initialized"); return OPAL_ERROR; } if( PMI_TRUE != initialized && PMI_SUCCESS != (rc = PMI_Init(&spawned)) ) { OPAL_PMI_ERROR(rc, "PMI_Init"); return OPAL_ERROR; } // Initialize space demands rc = PMI_KVS_Get_value_length_max(&pmix_vallen_max); if( PMI_SUCCESS != rc ) { OPAL_PMI_ERROR(rc, "PMI_KVS_Get_value_length_max"); goto err_exit; } rc = PMI_KVS_Get_name_length_max(&pmix_kvslen_max); if (PMI_SUCCESS != rc ) { OPAL_PMI_ERROR(rc, "PMI_KVS_Get_name_length_max"); goto err_exit; } rc = PMI_KVS_Get_key_length_max(&pmix_keylen_max); if( PMI_SUCCESS != rc ) { OPAL_PMI_ERROR(rc, "PMI_KVS_Get_key_length_max"); goto err_exit; } // Initialize job environment information pmix_id = (char*)malloc(pmix_vallen_max); if( pmix_id == NULL ){ ret = OPAL_ERR_OUT_OF_RESOURCE; goto err_exit; } /* Get domain id */ if (PMI_SUCCESS != (rc = PMI_Get_kvs_domain_id(pmix_id, pmix_vallen_max))) { free(pmix_id); goto err_exit; } /* Slurm PMI provides the job id as an integer followed * by a '.', followed by essentially a stepid. The first integer * defines an overall job number. The second integer is the number of * individual jobs we have run within that allocation. So we translate * this as the overall job number equating to our job family, and * the individual number equating to our local jobid */ jobfam = strtoul(pmix_id, &tmp, 10); if (NULL == tmp) { /* hmmm - no '.', so let's just use zero */ stepid = 0; } else { tmp++; /* step over the '.' */ stepid = strtoul(tmp, NULL, 10); } /* now build the jobid */ s1_jobid = (jobfam << 16) | stepid; free(pmix_id); /* get our rank */ ret = PMI_Get_rank(&s1_rank); if( PMI_SUCCESS != ret ) { OPAL_PMI_ERROR(ret, "PMI_Get_rank"); goto err_exit; } /* store our name in the opal_proc_t so that * debug messages will make sense - an upper * layer will eventually overwrite it, but that * won't do any harm */ s1_pname.jid = s1_jobid; s1_pname.vid = s1_rank; opal_proc_set_name((opal_process_name_t*)&s1_pname); opal_output_verbose(2, opal_pmix_base_framework.framework_output, "%s pmix:s1: assigned tmp name", OPAL_NAME_PRINT(*(opal_process_name_t*)&s1_pname)); pmix_kvs_name = (char*)malloc(pmix_kvslen_max); if( pmix_kvs_name == NULL ){ ret = OPAL_ERR_OUT_OF_RESOURCE; goto err_exit; } rc = PMI_KVS_Get_my_name(pmix_kvs_name, pmix_kvslen_max); if( PMI_SUCCESS != rc ) { OPAL_PMI_ERROR(rc, "PMI_KVS_Get_my_name"); goto err_exit; } /* get our local proc info to find our local rank */ if (PMI_SUCCESS != (rc = PMI_Get_clique_size(&s1_nlranks))) { OPAL_PMI_ERROR(rc, "PMI_Get_clique_size"); return rc; } /* now get the specific ranks */ s1_lranks = (int*)calloc(s1_nlranks, sizeof(int)); if (NULL == s1_lranks) { rc = OPAL_ERR_OUT_OF_RESOURCE; OPAL_ERROR_LOG(rc); return rc; } if (PMI_SUCCESS != (rc = PMI_Get_clique_ranks(s1_lranks, s1_nlranks))) { OPAL_PMI_ERROR(rc, "PMI_Get_clique_ranks"); free(s1_lranks); return rc; } /* find ourselves */ for (i=0; i < s1_nlranks; i++) { if (s1_rank == s1_lranks[i]) { s1_lrank = i; s1_nrank = i; break; } } /* get universe size */ ret = PMI_Get_universe_size(&s1_usize); if (PMI_SUCCESS != ret) { OPAL_PMI_ERROR(ret, "PMI_Get_universe_size"); goto err_exit; } /* push this into the dstore for subsequent fetches */ OBJ_CONSTRUCT(&kv, opal_value_t); kv.key = strdup(OPAL_DSTORE_UNIV_SIZE); kv.type = OPAL_UINT32; kv.data.uint32 = s1_usize; if (OPAL_SUCCESS != (ret = opal_dstore.store(opal_dstore_internal, &OPAL_PROC_MY_NAME, &kv))) { OPAL_ERROR_LOG(ret); OBJ_DESTRUCT(&kv); goto err_exit; } OBJ_DESTRUCT(&kv); /* get job size */ ret = PMI_Get_size(&s1_jsize); if (PMI_SUCCESS != ret) { OPAL_PMI_ERROR(ret, "PMI_Get_size"); goto err_exit; } /* get appnum */ ret = PMI_Get_appnum(&s1_appnum); if (PMI_SUCCESS != ret) { OPAL_PMI_ERROR(ret, "PMI_Get_appnum"); goto err_exit; } return OPAL_SUCCESS; err_exit: PMI_Finalize(); return ret; }
bool mca_common_pmi_init (void) { if (0 < mca_common_pmi_init_count++) { return true; } #if WANT_PMI2_SUPPORT { int spawned, size, rank, appnum; int rc; size = -1; rank = -1; appnum = -1; /* deal with a Slurm bug by first checking if we were * even launched by a PMI server before attempting * to use PMI */ if (NULL == getenv("PMI_FD")) { mca_common_pmi_init_count--; return false; } /* if we can't startup PMI, we can't be used */ if (PMI2_Initialized ()) { return true; } if (PMI2_SUCCESS != (rc = PMI2_Init(&spawned, &size, &rank, &appnum))) { opal_show_help("help-common-pmi.txt", "pmi2-init-failed", true, rc); mca_common_pmi_init_count--; return false; } if (size < 0 || rank < 0 ) { opal_show_help("help-common-pmi.txt", "pmi2-init-returned-bad-values", true); mca_common_pmi_init_count--; return false; } mca_common_pmi_init_size = size; mca_common_pmi_init_rank = rank; mca_common_pmi_init_count--; return true; } #else { PMI_BOOL initialized; /* deal with a Slurm bug by first checking if we were * even launched by a PMI server before attempting * to use PMI */ if (NULL == getenv("PMI_FD")) { mca_common_pmi_init_count--; return false; } if (PMI_SUCCESS != PMI_Initialized(&initialized)) { mca_common_pmi_init_count--; return false; } if (PMI_TRUE != initialized && PMI_SUCCESS != PMI_Init(&initialized)) { mca_common_pmi_init_count--; return false; } } #endif return true; }
int main(int argc, char **argv, char **envp) { int pmi_rank = -1; int pmi_process_group_size = -1; int num_local_procs = 0; int *local_rank_ids = NULL; int spawned = PMI_FALSE; int rc = EXIT_SUCCESS; char *err = NULL; PMI_BOOL pmi_initialized = PMI_FALSE; int pmi_vallen_max, max_length; char *pmi_kvs_name; /* sanity */ if (PMI_SUCCESS != PMI_Initialized(&pmi_initialized) || PMI_TRUE == pmi_initialized) { fprintf(stderr, "=== ERROR: PMI sanity failure\n"); return EXIT_FAILURE; } if (PMI_SUCCESS != PMI_Init(&spawned)) { err = "PMI_Init failure!"; goto done; } if (PMI_SUCCESS != PMI_Get_size(&pmi_process_group_size)) { err = "PMI_Get_size failure!"; goto done; } if (PMI_SUCCESS != PMI_Get_rank(&pmi_rank)) { err = "PMI_Get_rank failure!"; goto done; } if (PMI_SUCCESS != PMI_Get_clique_size(&num_local_procs)) { err = "PMI_Get_clique_size failure!"; goto done; } if (PMI_SUCCESS != PMI_KVS_Get_value_length_max(&pmi_vallen_max)) { err = "PMI_KVS_Get_value_length_max failure!"; goto done; } if (PMI_SUCCESS != PMI_KVS_Get_name_length_max(&max_length)) { err = "PMI_KVS_Get_name_length_max failure!"; goto done; } pmi_kvs_name = (char*)malloc(max_length); if (NULL == pmi_kvs_name) { err = "malloc failure!"; goto done; } if (PMI_SUCCESS != PMI_KVS_Get_my_name(pmi_kvs_name,max_length)) { err = "PMI_KVS_Get_my_name failure!"; goto done; } if (NULL == (local_rank_ids = calloc(num_local_procs, sizeof(int)))) { err = "out of resources"; goto done; } if (PMI_SUCCESS != PMI_Get_clique_ranks(local_rank_ids, num_local_procs)) { err = "PMI_Get_clique_size failure!"; goto done; } /* lowest local rank will print env info and tag its output*/ // if (pmi_rank == local_rank_ids[0]) { // for (; NULL != envp && NULL != *envp; ++envp) { // printf("===[%d]: %s\n", pmi_rank, *envp); // } //} done: if (PMI_TRUE == pmi_initialized) { if (PMI_SUCCESS != PMI_Finalize()) { err = "PMI_Finalize failure!"; } } if (NULL != err) { fprintf(stderr, "=== ERROR [rank:%d] %s\n", pmi_rank, err); rc = EXIT_FAILURE; } return rc; }
static int mca_initialize_pmi_v1(void) { PMI_BOOL initialized; int spawned; int rc, ret = OPAL_ERROR; if (PMI_SUCCESS != (rc = PMI_Initialized(&initialized))) { OPAL_PMI_ERROR(rc, "PMI_Initialized"); return OPAL_ERROR; } if( PMI_TRUE != initialized && PMI_SUCCESS != (rc = PMI_Init(&spawned)) ) { OPAL_PMI_ERROR(rc, "PMI_Init"); return OPAL_ERROR; } // Initialize space demands rc = PMI_KVS_Get_value_length_max(&pmi_vallen_max); if( PMI_SUCCESS != rc ) { OPAL_PMI_ERROR(rc, "PMI_KVS_Get_value_length_max"); goto err_exit; } rc = PMI_KVS_Get_name_length_max(&pmi_kvslen_max); if (PMI_SUCCESS != rc ) { OPAL_PMI_ERROR(rc, "PMI_KVS_Get_name_length_max"); goto err_exit; } rc = PMI_KVS_Get_key_length_max(&pmi_keylen_max); if( PMI_SUCCESS != rc ) { OPAL_PMI_ERROR(rc, "PMI_KVS_Get_key_length_max"); goto err_exit; } // Initialize job environment information rc = PMI_Get_rank(&pmi_rank); if( PMI_SUCCESS != rc ) { OPAL_PMI_ERROR(rc, "PMI_Get_rank"); return OPAL_ERROR; } rc = PMI_Get_universe_size(&pmi_usize); if( PMI_SUCCESS != rc ) { OPAL_PMI_ERROR(rc, "PMI_Get_universe_size"); goto err_exit; } rc = PMI_Get_size(&pmi_size); if( PMI_SUCCESS != rc ) { OPAL_PMI_ERROR(rc, "PMI_Get_size"); goto err_exit; } rc = PMI_Get_appnum(&pmi_appnum); if( PMI_SUCCESS != rc ) { OPAL_PMI_ERROR(rc, "PMI_Get_appnum"); goto err_exit; } pmi_kvs_name = (char*)malloc(pmi_kvslen_max); if( pmi_kvs_name == NULL ){ ret = OPAL_ERR_OUT_OF_RESOURCE; goto err_exit; } rc = PMI_KVS_Get_my_name(pmi_kvs_name,pmi_kvslen_max); if( PMI_SUCCESS != rc ) { OPAL_PMI_ERROR(rc, "PMI2_Job_GetId"); goto err_exit; } return OPAL_SUCCESS; err_exit: PMI_Finalize(); return ret; }